Industrial Grade OCR

We have all run across OCR mistakes and reported them as typos. It gets me wondering about the OCR capabilities of FaithLife or Kindle or other electronic book publishers.

Are industrial OCR capabilities significantly better than the OCR programs available to us?

Find more posts tagged with

Comments

Justin Gatlin

So OCR is actually a very difficult problem. There is not a huge distinction at this point between software levels (although machine learning is exciting on this front). I believe it is a time/accuracy balance. A computer can be configured to make a closer analysis for higher accuracy, but there is a point of diminishing returns. Should Logos spend twice as long to get a text which is 1/2 % more accurate?

scooter

Hi, John: The world is full of acronyms. I did not know what OCR meant till I looked up the article Justin had included in his post.

Mark Barnes

Are industrial OCR capabilities significantly better than the OCR programs available to us?

The software isn't much different, but the hardware can be: https://www.logos.com/features/bookscanner. Software that's for a specific hardware setup is usually more accurate, because the software can compensate more exactly for page curvature etc. if it knows exactly how things were scanned.

Many of the older resources with OCR errors will have been done on the set up linked above.

John Connell

https://community.logos.com/discussion/comment/821005#Comment_821005

Thank you Justin and Mark. Exactly the information I was seeking.

Kyle G. Anderson

https://community.logos.com/discussion/comment/821019#Comment_821019

FWIW. I should probably look at getting that page pulled since it's no longer relevant.

For approximately the last three to four years we, generally speaking, convert text one of two ways:

For resources that need do not exist as an exportable file, we don't use OCR. We double key the book and run a diff and then consult with the print.
For resources that exist as an exportable file we run a diff on the final files against the original exported files.

In my experience resources with a higher proportion of typos tend to be resources made in 2011 or earlier.

John Connell

https://community.logos.com/discussion/comment/821044#Comment_821044

Kyle,

Back confused again. What does it mean to "double key the book?"

MJ. Smith

https://community.logos.com/discussion/comment/821126#Comment_821126

Two people enter the text via keyboard; there results are compared.

Colin

https://community.logos.com/discussion/comment/821132#Comment_821132

Two people enter the text via keyboard; there results are compared.

Two people enter the text via keyboard; their results are compared.

Ooh MJ, I guessed you were using a deliberate typo to illustrate the process [;)]

John Connell

https://community.logos.com/discussion/comment/821136#Comment_821136

Wow, I had no idea people were retyping these resources rather than using OCR. This is totally surprising to me.

Mark Barnes

https://community.logos.com/discussion/comment/821137#Comment_821137

Wow, I had no idea people were retyping these resources rather than using OCR. This is totally surprising to me.

It used to be OCR, so some older resources (the ones with lots of typos) will have been done with OCR.

Kyle G. Anderson

https://community.logos.com/discussion/comment/821140#Comment_821140

Wow, I had no idea people were retyping these resources rather than using OCR. This is totally surprising to me.

It used to be OCR, so some older resources (the ones with lots of typos) will have been done with OCR.

Correct. As advanced as OCR is, it still doesn't do well on really old PDFs (which often what we have). For example it can easily mistake "e" as "c" or a "cl" as a "d". This isn't surprising. Sometimes it's even hard for a human to tell the difference and this will be with the best print available.

We found we got better results at a better price double keying the resource and running a diff than doing an OCR and proofreading.

Dave Thawley

https://community.logos.com/discussion/comment/821140#Comment_821140

Wow, I had no idea people were retyping these resources rather than using OCR. This is totally surprising to me.

It used to be OCR, so some older resources (the ones with lots of typos) will have been done with OCR.

I was part of a team once contracted to OCR a couple hundred old documents. After a while we found there were so many errors we ended up typing a lot of it in again. It took hundreds of hours when we were expecting in to take a week

Paul

https://community.logos.com/discussion/comment/821431#Comment_821431

Thanks for the explanation Kylie - it makes me more understanding of the difficulties involved in producing some of the books. It also illustrates that its hard work!

Jerry Thompson

Can anyone tell me how to produce an iota subscript with the Logos Greek Keyboard? I'm didn't want to interrupt this forum thread, but I can't find anywhere to ask the question? I felt sure someone among you would know the answer. Thanks, JT

Graham Criddle

https://community.logos.com/discussion/comment/823349#Comment_823349

Hi Jerry - and welcome to the forums

Can anyone tell me how to produce an iota subscript with the Logos Greek Keyboard? I'm didn't want to interrupt this forum thread, but I can't find anywhere to ask the question? I felt sure someone among you would know the answer. Thanks, JT

I would cross-post this in the Logos 6 forum - https://community.logos.com/forums/124.aspx

Graham